Hybrid Deep Transfer Learning Framework for Enhanced Brain Tumor Classification Using MRI

Hybrid Deep Transfer Learning Framework for Enhanced Brain Tumor Classification Using MRI

Brain tumors are a major global health concern, according to the report by the World Health Organization. Among others, the methods that help in diagnosis include MRI and CT scans. However, MRI is still preferred due to its detailed non-invasive imaging capability. To improve the accuracy and efficiency of detecting and classifying tumors, a new hybrid deep transfer learning framework has been proposed to automate the categorization. This framework uses two MRI datasets that were preprocessed using data augmentation. The other component integrates five deep learning, pre-trained models: DenseNet121, Xception, ResNet50, MobileNetV2, and Inception V3, with different layers added and their performance enhanced by transfer learning. Through these models, deep features are extracted in MRI images for training. The capabilities of the suggested framework will then be assessed in terms of accuracy, recall, precision, and F1 score. The experiment results show that the hybrid Xception model has been much more effective in distinguishing glioma, meningioma, no tumor, and pituitary tumors, among others. It recorded accuracies ranging from 98.60% to 99.70% for Dataset-I and 98.20% to 98.60% for Dataset-II across different categories. According to the results obtained, it seems that the hybrid Xception model holds potential for improving medical diagnosis through accurate classification of brain tumors.